/FC-AIDE-Keras

FC-AIDE: Fully Convolutional Pixel Adaptive Image Denoiser

Primary LanguagePythonMIT LicenseMIT

FC-AIDE: Fully Convolutional Adaptive Image Denoiser

Contents

  • code for testing

    • 'sigma_estimation.py'
      : estimate a noise sigma of the noisy image
    • 'test_fc_aide_sup.py'
      : denoise a noisy image using the supervised model trained on the specific noise level
    • 'test_fc_aide_ft.py'
      : denoise a noisy image using the supervised model + fine-tuning
    • 'test_fc_aide_blind_sup.py'
      : denoise a noisy image using the supervised model trained on various noise level
    • 'test_fc_aide_blind_ft.py'
      : denoise a noisy image using the supervised model + fine-tuning
    • 'test_fc_aide_blind_estimated_sigma_ft.py'
      : estimate a sigma of the noise level and then denoise using the supervised model + fine-tuning

  • weights

    • 'sigmaX.hdf5' : a weight trained on a specific noise level([15, 25, 30, 50, 75])
    • 'blind.hdf5' : the weight trained on various noise levels([0,50])

  • test images

    • 'Set13'
    • 'BSD68'
    • 'Medical60'

Results

Set13

The average PSNR(dB) on Set13.

Sigma BM3D RED MemNet DnCNN-S DnCNN-B FC-AIDE_S FC-AIDE_S+FT FC-AIDE_B FC-AIDE_B+FT
15 31.98 - - 32.21 31.58 32.08 32.59 31.72 32.52
25 29.44 - - 29.63 29.22 29.57 30.14 29.34 30.03
30 28.56 28.91 28.83 28.64 28.41 28.73 29.28 28.50 29.19
50 26.05 26.28 26.39 26.09 26.07 26.24 26.87 26.05 26.77
75 24.16 - - 24.03 18.33 24.24 24.97 21.07 24.89

BSD68

The average PSNR(dB) on BSD68.

Sigma BM3D RED MemNet DnCNN-S DnCNN-B FC-AIDE_S FC-AIDE_S+FT FC-AIDE_B FC-AIDE_B+FT
15 31.07 - - 31.72 31.60 31.63 31.75 31.47 31.72
25 28.56 - - 29.22 29.15 29.18 29.31 29.04 29.26
30 27.74 28.45 28.42 28.35 28.34 28.35 28.49 28.24 28.44
50 25.60 26.29 26.34 26.21 26.20 26.24 26.38 26.12 26.33
75 24.19 - - 24.62 18.68 24.74 24.87 21.42 24.76

Mismatch / blind case

Requirements and Dependencies

  • Python 2.7 / Python 3.6
  • CUDA v8.0 / CuDNN v5.1
  • Tensorflow v1.2.1
  • Keras v2.0.8

citation

@ARTICLE{2018arXiv180707569C,  
    author = {{Cha}, S. and {Moon}, T.}a,  
    title = "{Fully Convolutional Pixel Adaptive Image Denoiser}",  
    journal = {ArXiv e-prints},  
    archivePrefix = "arXiv",  
    eprint = {1807.07569},  
    primaryClass = "cs.CV",  
    keywords = {Computer Science - Computer Vision and Pattern Recognition, Computer Science - Machine Learning, Statistics - Machine Learning},  
    year = 2018,  
    month = jul,  
    adsurl = {http://adsabs.harvard.edu/abs/2018arXiv180707569C},  
    adsnote = {Provided by the SAO/NASA Astrophysics Data System}  
}